Lies, Damn Lies, and Data Visualization
Predictive Analytics Key Component of Customer Experience Management
Predictive Analytics in Higher Education
Trends in Computational Healthcare Research and Digital Therapeutics
Increasing Your Bottom Line with Predictive Analytics
Rich Jolly, VP Analytics & Strategy, Avamere Family of Companies
Data Lake: Building a Bridge between Technology and Business
Shaun Rankin, Ex SVP Data Management, Citizens Bank
Leveraging Predictive Analytics to Power Customer Data &...
Dean Abbott, Co-founder & Chief Data Scientist, SmarterHQ
Thank you for Subscribing to CIO Applications Weekly Brief
Predictive Analytics in Healthcare: It's Not Happening
By Kirk Kirksey, VP & CIO, UT Southwestern Medical Center
To state the obvious, ‘we are not there yet’. Not only are we not ‘There’ yet, we remain a long, long way from ‘There’ and we don’t seem to be getting any closer to ‘There’. Dashboards and their hybrids have become synonyms for Healthcare Analytics, and when we buy Healthcare Analytics we know we are buying dash boards or their kin. It’s time to move to that Next Plateau— predictive analytics.
The Rise and Stall of the Dashboard
Dashboards are flashy. Dashboards are cool. Dashboards can integrate data from all those disparate systems we need to operate our hospitals and clinics. Dashboards can scoop data up in real time. Dashboards can even use “Big Data”. There is no doubt. Dashboards are an invaluable management tool, and today we don’t know how we ever got along without them. But like everything else, dashboards have strengths and weaknesses.
A well-designed dashboard can certainly highlight problems, but it is up to an experienced manager to evaluate dashboard information, then formulate solutions and take action. So, as adept as dashboards have become at presenting data—real time, disparate, big, and otherwise—most dash boards remain sophisticated management reports. I make this observation because the majority of dashboards I see are not designed to uncover the root cause of a problem; only the symptoms of cause. Root cause and solutions are someone else’s job. Uncovering cause and causal relationships, on the other hand, is the prime commodity of predictive analytics.
Intervention—The Reason for Prediction
I have two jobs. I’m an academic, and I am a healthcare CIO. In my role as an academic, my colleagues and I publish predictive analytic models based on healthcare data. Publishing helps my academic career. Published articles look good on my resume. Sure, it would be nice if someone someday would use this work to improve healthcare, but chances are slim to none this will happen.
On the other hand, as a healthcare CIO I am interested in action. I want to help devise interventions leading to higher quality healthcare, lower costs, and improved patient outcomes. I want these interventions to be evidence based—derived from sound predictive models calculated using reliable data.
Consider this proposition as an example: “The use of a patient portal by congestive heart failure patients (CHF) will lead to improved outcomes”.
Predictive analytics, using sound statistical and sampling techniques for minimizing risk, must form the foundation of an intervention
We (vendors and healthcare CIOs) have been peddling this rhetorical morsel since Day One. But in truth this intuitive assumption, though widely accepted, has never been proven. We have no idea if using a patient portal really has any effect whatsoever on patient outcomes. Think of the possibilities if we could prove a positive causal effect between portal use and improved outcomes. We could create interventions designed to foster more effective use of portals by patients. We could design targeted training programs. We could mount patient recruitment programs. We might even determine giving computers and mobile devices to CHF patients is cost effective because portal use reduces emergency room visits.
Before getting too excited about interventions there something you should know. Interventions can be dangerous and expensive. Enter predictive analysis. Predictive analytics, using sound statistical and sampling techniques for minimizing risk, must form the foundation of an intervention. Relying on intuition and unfounded assumptions, no matter how obvious, is simply too risky. Let me demonstrate the importance of this notion with a non-healthcare example—the self-driving car.
The self-driving car is a multi-ton predictive analytics processor on wheels. When algorithms inside the car’s computer predict the vehicle will hit a pedestrian unless an intervention occurs. The car engages the brakes (intervention) as the result of a proven predictive model. Recently we have seen how disaster (and even death) can occur when an intervention (or lack of intervention) is based on faulty analytics. These life-and-death stakes are just as high in healthcare, and interventions derived from faith-based assumptions can be deadly.
Getting to the Next Plateau
My naiveté may be showing, but I believe we will eventually reach the Predictive Analytics plateau in healthcare. But the road is winding and complex. In my opinion, the trigger we need will be value-based healthcare (secret code for risk based contracting). Sharing risk successfully means understanding costs, and understanding healthcare costs requires lots and lots of data. Once these risk sharing contracts are in place, managing populations of patients and networks of providers, will be even more data intensive. As a result of this trend, wise healthcare organizations and vendors are creating analytics technology architectures. These technologies will be used to first produce, dare I say it, dashboards. But necessary analytics components like warehouses, data normalization structures, data governance techniques, data lakes and more are being developed procured, and implemented. With cloud-based technologies, the analytics price tag is looking better and better. Bottom line—we are starting to see the emergence of cost effective technology architectures capable of producing meaningful predictive models. Once in place, these will be the platforms for predictive analytics. But when it comes to predictive analytics, technology is not nearly enough.
Having people with the skills capable of producing sound predictive models is the key. Today these skilled professionals are rare, and this skills gap will be our biggest hurdle in the foreseeable future.
Other than academic medical centers we rarely see these types of individuals in a typical healthcare delivery organization. This is not easy stuff to teach or learn. First and foremost, this new workforce must be well schooled in the scientific method. Practitioners must be able to devise testable questions (scientists would call this ‘forming a hypothesis’). Once a hypothesis is formed it must be rigorously tested with sound statistical techniques. A thorough understanding of the data—its strengths, weaknesses, inclusions and exclusions—is an absolute requirement. Data integration, normalization, technology architectures, data governance and more all go into the calculus of predictive analytics. I could go on for hours, but the bottom line is obvious. We must learn and teach a new kind of workforce if we are to devise evidence base interventions capable of changing healthcare for the better.
Without interventions created by sound predictive analytics we will never fully realize the value of healthcare data. Technology is not the problem, and with the proliferation of EMRs we have plenty of data. We know how to train qualified professionals, but sadly the process of creating this new workforce is slow-moving and embryonic. Are we ready to travel beyond dashboards, and move to predictive analytics in healthcare? Sadly, appears the answer to this question is a resounding ‘Not Yet’.